Porn To Run: Online Advertising Gets A Flesh Blocker

Sex sells. As long as you don't go too far.

Here is your typical family on the beach. Two boys playing in the sand. Husband and wife looking on. She's got only about about 5 percent of her body covered in clothing. Exposed flesh reigns.

Is this porn? No.

At the side here is a woman in lingerie, with plenty of cleavage showing. Is this porn? Probably not. Maybe to some viewers. You decide.

Down below is a risque ad. Young girls rough-housing in panties, or less. Hard to tell.

Porn? Yes.

At least that's the verdict of a new tool from the Rubicon Project called Helix, which scrutinizes any image and classifies it as porn or non-porn.

It's a tool designed to protect online advertisers and Web site publishers from the rapid explosion in and vagaries of ad networks. The networks push to fill all unsold ad space on the Web -- of which there is plenty -- and that can lead to some visual and visible marketing clashes.

As AdWeek reported,  Procter & Gamble likely didn't intend for preroll ads for its Tide, Cover Girl and Herbal Essences brands to show up just before Playboy playmate Kendra Wilkinson flashed a breast at onlookers on TMZ.com. Global bankers HSBC and American Express probably did not intend to run banner ads "within clicking distance" of nude photos of High School Musical star Vanessa Hudgens on the celebrity-gossip blog What Would Tyler Durden Do? (WWTDD.com).

So now, P&G, HSBC and Amex can get some relief, from a flesh blocker like Helix. The scanning software tries to identify porn, automatically, and bring them to the attention of humans, before they appear on a site. If a publisher wants to keep P&G as a client, this will flag pornographic ads before they appear on the same page as a Tide ad.

It's an unsexy, you might say, application of artificial intelligence. Helix basically has taught itself what is porn and what is not porn, according to Rubicon Project CEO and co-founder Frank Addante.

Humans fed 200,000 pieces of porn into a database underlying Helix about three months ago. And tens of thousands of images that were designated as "not porn" into another database.

Helix extracted key features of each image fed into Rubicon's servers, creating "point clouds" or sets of three-dimensional coordinates. Spectral analysis also is applied to the lighting and color characteristics of each image.

The result: A 'feature vector' of about 2,000 floating point numbers that define the content of the image.

Those numbers are then fed into a forest of decision trees that, based on the training set of images, decides whether each new image should be classified as porn or not porn. Helix determines when a lot of flesh is designed to arouse attention (and that other thing) and when it is not. It can be not just the amount, but the positions, the angles, or the lighting involved.

But Helix does not act alone in blocking suggestive ads from appearing on a site. The manager of the site makes the final decision. Helix just tees up the choice.

Rubicon, with this service, hopes to protect the growth of online advertising. Online ad growth is a bright spot, relatively speaking, in this contracting economy. Web site publishers still want to hit the $106 billion mark in online advertising worldwide by 2011 that IDC was predicting in June.  But online advertisers have been cutting back, as one October survey not unexpectedly found.

If not enough publishers sign up for the service, of course, there's always one other potentially profitable play for this project.

Sell it to spouses, for installation on computers at home.

You can't tame all of the Internet. Just pieces.